| 1 |
/* Copyright (C) 2013-2015 Free Software Foundation, Inc. |
| 2 |
|
| 3 |
This file is part of GCC. |
| 4 |
|
| 5 |
GCC is free software; you can redistribute it and/or modify |
| 6 |
it under the terms of the GNU General Public License as published by |
| 7 |
the Free Software Foundation; either version 3, or (at your option) |
| 8 |
any later version. |
| 9 |
|
| 10 |
GCC is distributed in the hope that it will be useful, |
| 11 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 |
GNU General Public License for more details. |
| 14 |
|
| 15 |
Under Section 7 of GPL version 3, you are granted additional |
| 16 |
permissions described in the GCC Runtime Library Exception, version |
| 17 |
3.1, as published by the Free Software Foundation. |
| 18 |
|
| 19 |
You should have received a copy of the GNU General Public License and |
| 20 |
a copy of the GCC Runtime Library Exception along with this program; |
| 21 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 22 |
<http://www.gnu.org/licenses/>. */ |
| 23 |
|
| 24 |
#ifndef _IMMINTRIN_H_INCLUDED |
| 25 |
#error "Never use <avx512erintrin.h> directly; include <immintrin.h> instead." |
| 26 |
#endif |
| 27 |
|
| 28 |
#ifndef _AVX512ERINTRIN_H_INCLUDED |
| 29 |
#define _AVX512ERINTRIN_H_INCLUDED |
| 30 |
|
| 31 |
#ifndef __AVX512ER__ |
| 32 |
#pragma GCC push_options |
| 33 |
#pragma GCC target("avx512er") |
| 34 |
#define __DISABLE_AVX512ER__ |
| 35 |
#endif /* __AVX512ER__ */ |
| 36 |
|
| 37 |
/* Internal data types for implementing the intrinsics. */ |
| 38 |
typedef double __v8df __attribute__ ((__vector_size__ (64))); |
| 39 |
typedef float __v16sf __attribute__ ((__vector_size__ (64))); |
| 40 |
|
| 41 |
/* The Intel API is flexible enough that we must allow aliasing with other |
| 42 |
vector types, and their scalar components. */ |
| 43 |
typedef float __m512 __attribute__ ((__vector_size__ (64), __may_alias__)); |
| 44 |
typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__)); |
| 45 |
|
| 46 |
typedef unsigned char __mmask8; |
| 47 |
typedef unsigned short __mmask16; |
| 48 |
|
| 49 |
#ifdef __OPTIMIZE__ |
| 50 |
extern __inline __m512d |
| 51 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 52 |
_mm512_exp2a23_round_pd (__m512d __A, int __R) |
| 53 |
{ |
| 54 |
__m512d __W; |
| 55 |
return (__m512d) __builtin_ia32_exp2pd_mask ((__v8df) __A, |
| 56 |
(__v8df) __W, |
| 57 |
(__mmask8) -1, __R); |
| 58 |
} |
| 59 |
|
| 60 |
extern __inline __m512d |
| 61 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 62 |
_mm512_mask_exp2a23_round_pd (__m512d __W, __mmask8 __U, __m512d __A, int __R) |
| 63 |
{ |
| 64 |
return (__m512d) __builtin_ia32_exp2pd_mask ((__v8df) __A, |
| 65 |
(__v8df) __W, |
| 66 |
(__mmask8) __U, __R); |
| 67 |
} |
| 68 |
|
| 69 |
extern __inline __m512d |
| 70 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 71 |
_mm512_maskz_exp2a23_round_pd (__mmask8 __U, __m512d __A, int __R) |
| 72 |
{ |
| 73 |
return (__m512d) __builtin_ia32_exp2pd_mask ((__v8df) __A, |
| 74 |
(__v8df) _mm512_setzero_pd (), |
| 75 |
(__mmask8) __U, __R); |
| 76 |
} |
| 77 |
|
| 78 |
extern __inline __m512 |
| 79 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 80 |
_mm512_exp2a23_round_ps (__m512 __A, int __R) |
| 81 |
{ |
| 82 |
__m512 __W; |
| 83 |
return (__m512) __builtin_ia32_exp2ps_mask ((__v16sf) __A, |
| 84 |
(__v16sf) __W, |
| 85 |
(__mmask16) -1, __R); |
| 86 |
} |
| 87 |
|
| 88 |
extern __inline __m512 |
| 89 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 90 |
_mm512_mask_exp2a23_round_ps (__m512 __W, __mmask16 __U, __m512 __A, int __R) |
| 91 |
{ |
| 92 |
return (__m512) __builtin_ia32_exp2ps_mask ((__v16sf) __A, |
| 93 |
(__v16sf) __W, |
| 94 |
(__mmask16) __U, __R); |
| 95 |
} |
| 96 |
|
| 97 |
extern __inline __m512 |
| 98 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 99 |
_mm512_maskz_exp2a23_round_ps (__mmask16 __U, __m512 __A, int __R) |
| 100 |
{ |
| 101 |
return (__m512) __builtin_ia32_exp2ps_mask ((__v16sf) __A, |
| 102 |
(__v16sf) _mm512_setzero_ps (), |
| 103 |
(__mmask16) __U, __R); |
| 104 |
} |
| 105 |
|
| 106 |
extern __inline __m512d |
| 107 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 108 |
_mm512_rcp28_round_pd (__m512d __A, int __R) |
| 109 |
{ |
| 110 |
__m512d __W; |
| 111 |
return (__m512d) __builtin_ia32_rcp28pd_mask ((__v8df) __A, |
| 112 |
(__v8df) __W, |
| 113 |
(__mmask8) -1, __R); |
| 114 |
} |
| 115 |
|
| 116 |
extern __inline __m512d |
| 117 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 118 |
_mm512_mask_rcp28_round_pd (__m512d __W, __mmask8 __U, __m512d __A, int __R) |
| 119 |
{ |
| 120 |
return (__m512d) __builtin_ia32_rcp28pd_mask ((__v8df) __A, |
| 121 |
(__v8df) __W, |
| 122 |
(__mmask8) __U, __R); |
| 123 |
} |
| 124 |
|
| 125 |
extern __inline __m512d |
| 126 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 127 |
_mm512_maskz_rcp28_round_pd (__mmask8 __U, __m512d __A, int __R) |
| 128 |
{ |
| 129 |
return (__m512d) __builtin_ia32_rcp28pd_mask ((__v8df) __A, |
| 130 |
(__v8df) _mm512_setzero_pd (), |
| 131 |
(__mmask8) __U, __R); |
| 132 |
} |
| 133 |
|
| 134 |
extern __inline __m512 |
| 135 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 136 |
_mm512_rcp28_round_ps (__m512 __A, int __R) |
| 137 |
{ |
| 138 |
__m512 __W; |
| 139 |
return (__m512) __builtin_ia32_rcp28ps_mask ((__v16sf) __A, |
| 140 |
(__v16sf) __W, |
| 141 |
(__mmask16) -1, __R); |
| 142 |
} |
| 143 |
|
| 144 |
extern __inline __m512 |
| 145 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 146 |
_mm512_mask_rcp28_round_ps (__m512 __W, __mmask16 __U, __m512 __A, int __R) |
| 147 |
{ |
| 148 |
return (__m512) __builtin_ia32_rcp28ps_mask ((__v16sf) __A, |
| 149 |
(__v16sf) __W, |
| 150 |
(__mmask16) __U, __R); |
| 151 |
} |
| 152 |
|
| 153 |
extern __inline __m512 |
| 154 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 155 |
_mm512_maskz_rcp28_round_ps (__mmask16 __U, __m512 __A, int __R) |
| 156 |
{ |
| 157 |
return (__m512) __builtin_ia32_rcp28ps_mask ((__v16sf) __A, |
| 158 |
(__v16sf) _mm512_setzero_ps (), |
| 159 |
(__mmask16) __U, __R); |
| 160 |
} |
| 161 |
|
| 162 |
extern __inline __m128d |
| 163 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 164 |
_mm_rcp28_round_sd (__m128d __A, __m128d __B, int __R) |
| 165 |
{ |
| 166 |
return (__m128d) __builtin_ia32_rcp28sd_round ((__v2df) __B, |
| 167 |
(__v2df) __A, |
| 168 |
__R); |
| 169 |
} |
| 170 |
|
| 171 |
extern __inline __m128 |
| 172 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 173 |
_mm_rcp28_round_ss (__m128 __A, __m128 __B, int __R) |
| 174 |
{ |
| 175 |
return (__m128) __builtin_ia32_rcp28ss_round ((__v4sf) __B, |
| 176 |
(__v4sf) __A, |
| 177 |
__R); |
| 178 |
} |
| 179 |
|
| 180 |
extern __inline __m512d |
| 181 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 182 |
_mm512_rsqrt28_round_pd (__m512d __A, int __R) |
| 183 |
{ |
| 184 |
__m512d __W; |
| 185 |
return (__m512d) __builtin_ia32_rsqrt28pd_mask ((__v8df) __A, |
| 186 |
(__v8df) __W, |
| 187 |
(__mmask8) -1, __R); |
| 188 |
} |
| 189 |
|
| 190 |
extern __inline __m512d |
| 191 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 192 |
_mm512_mask_rsqrt28_round_pd (__m512d __W, __mmask8 __U, __m512d __A, int __R) |
| 193 |
{ |
| 194 |
return (__m512d) __builtin_ia32_rsqrt28pd_mask ((__v8df) __A, |
| 195 |
(__v8df) __W, |
| 196 |
(__mmask8) __U, __R); |
| 197 |
} |
| 198 |
|
| 199 |
extern __inline __m512d |
| 200 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 201 |
_mm512_maskz_rsqrt28_round_pd (__mmask8 __U, __m512d __A, int __R) |
| 202 |
{ |
| 203 |
return (__m512d) __builtin_ia32_rsqrt28pd_mask ((__v8df) __A, |
| 204 |
(__v8df) _mm512_setzero_pd (), |
| 205 |
(__mmask8) __U, __R); |
| 206 |
} |
| 207 |
|
| 208 |
extern __inline __m512 |
| 209 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 210 |
_mm512_rsqrt28_round_ps (__m512 __A, int __R) |
| 211 |
{ |
| 212 |
__m512 __W; |
| 213 |
return (__m512) __builtin_ia32_rsqrt28ps_mask ((__v16sf) __A, |
| 214 |
(__v16sf) __W, |
| 215 |
(__mmask16) -1, __R); |
| 216 |
} |
| 217 |
|
| 218 |
extern __inline __m512 |
| 219 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 220 |
_mm512_mask_rsqrt28_round_ps (__m512 __W, __mmask16 __U, __m512 __A, int __R) |
| 221 |
{ |
| 222 |
return (__m512) __builtin_ia32_rsqrt28ps_mask ((__v16sf) __A, |
| 223 |
(__v16sf) __W, |
| 224 |
(__mmask16) __U, __R); |
| 225 |
} |
| 226 |
|
| 227 |
extern __inline __m512 |
| 228 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 229 |
_mm512_maskz_rsqrt28_round_ps (__mmask16 __U, __m512 __A, int __R) |
| 230 |
{ |
| 231 |
return (__m512) __builtin_ia32_rsqrt28ps_mask ((__v16sf) __A, |
| 232 |
(__v16sf) _mm512_setzero_ps (), |
| 233 |
(__mmask16) __U, __R); |
| 234 |
} |
| 235 |
|
| 236 |
extern __inline __m128d |
| 237 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 238 |
_mm_rsqrt28_round_sd (__m128d __A, __m128d __B, int __R) |
| 239 |
{ |
| 240 |
return (__m128d) __builtin_ia32_rsqrt28sd_round ((__v2df) __B, |
| 241 |
(__v2df) __A, |
| 242 |
__R); |
| 243 |
} |
| 244 |
|
| 245 |
extern __inline __m128 |
| 246 |
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) |
| 247 |
_mm_rsqrt28_round_ss (__m128 __A, __m128 __B, int __R) |
| 248 |
{ |
| 249 |
return (__m128) __builtin_ia32_rsqrt28ss_round ((__v4sf) __B, |
| 250 |
(__v4sf) __A, |
| 251 |
__R); |
| 252 |
} |
| 253 |
|
| 254 |
#else |
| 255 |
#define _mm512_exp2a23_round_pd(A, C) \ |
| 256 |
__builtin_ia32_exp2pd_mask(A, (__v8df)_mm512_setzero_pd(), -1, C) |
| 257 |
|
| 258 |
#define _mm512_mask_exp2a23_round_pd(W, U, A, C) \ |
| 259 |
__builtin_ia32_exp2pd_mask(A, W, U, C) |
| 260 |
|
| 261 |
#define _mm512_maskz_exp2a23_round_pd(U, A, C) \ |
| 262 |
__builtin_ia32_exp2pd_mask(A, (__v8df)_mm512_setzero_pd(), U, C) |
| 263 |
|
| 264 |
#define _mm512_exp2a23_round_ps(A, C) \ |
| 265 |
__builtin_ia32_exp2ps_mask(A, (__v16sf)_mm512_setzero_ps(), -1, C) |
| 266 |
|
| 267 |
#define _mm512_mask_exp2a23_round_ps(W, U, A, C) \ |
| 268 |
__builtin_ia32_exp2ps_mask(A, W, U, C) |
| 269 |
|
| 270 |
#define _mm512_maskz_exp2a23_round_ps(U, A, C) \ |
| 271 |
__builtin_ia32_exp2ps_mask(A, (__v16sf)_mm512_setzero_ps(), U, C) |
| 272 |
|
| 273 |
#define _mm512_rcp28_round_pd(A, C) \ |
| 274 |
__builtin_ia32_rcp28pd_mask(A, (__v8df)_mm512_setzero_pd(), -1, C) |
| 275 |
|
| 276 |
#define _mm512_mask_rcp28_round_pd(W, U, A, C) \ |
| 277 |
__builtin_ia32_rcp28pd_mask(A, W, U, C) |
| 278 |
|
| 279 |
#define _mm512_maskz_rcp28_round_pd(U, A, C) \ |
| 280 |
__builtin_ia32_rcp28pd_mask(A, (__v8df)_mm512_setzero_pd(), U, C) |
| 281 |
|
| 282 |
#define _mm512_rcp28_round_ps(A, C) \ |
| 283 |
__builtin_ia32_rcp28ps_mask(A, (__v16sf)_mm512_setzero_ps(), -1, C) |
| 284 |
|
| 285 |
#define _mm512_mask_rcp28_round_ps(W, U, A, C) \ |
| 286 |
__builtin_ia32_rcp28ps_mask(A, W, U, C) |
| 287 |
|
| 288 |
#define _mm512_maskz_rcp28_round_ps(U, A, C) \ |
| 289 |
__builtin_ia32_rcp28ps_mask(A, (__v16sf)_mm512_setzero_ps(), U, C) |
| 290 |
|
| 291 |
#define _mm512_rsqrt28_round_pd(A, C) \ |
| 292 |
__builtin_ia32_rsqrt28pd_mask(A, (__v8df)_mm512_setzero_pd(), -1, C) |
| 293 |
|
| 294 |
#define _mm512_mask_rsqrt28_round_pd(W, U, A, C) \ |
| 295 |
__builtin_ia32_rsqrt28pd_mask(A, W, U, C) |
| 296 |
|
| 297 |
#define _mm512_maskz_rsqrt28_round_pd(U, A, C) \ |
| 298 |
__builtin_ia32_rsqrt28pd_mask(A, (__v8df)_mm512_setzero_pd(), U, C) |
| 299 |
|
| 300 |
#define _mm512_rsqrt28_round_ps(A, C) \ |
| 301 |
__builtin_ia32_rsqrt28ps_mask(A, (__v16sf)_mm512_setzero_ps(), -1, C) |
| 302 |
|
| 303 |
#define _mm512_mask_rsqrt28_round_ps(W, U, A, C) \ |
| 304 |
__builtin_ia32_rsqrt28ps_mask(A, W, U, C) |
| 305 |
|
| 306 |
#define _mm512_maskz_rsqrt28_round_ps(U, A, C) \ |
| 307 |
__builtin_ia32_rsqrt28ps_mask(A, (__v16sf)_mm512_setzero_ps(), U, C) |
| 308 |
|
| 309 |
#define _mm_rcp28_round_sd(A, B, R) \ |
| 310 |
__builtin_ia32_rcp28sd_round(A, B, R) |
| 311 |
|
| 312 |
#define _mm_rcp28_round_ss(A, B, R) \ |
| 313 |
__builtin_ia32_rcp28ss_round(A, B, R) |
| 314 |
|
| 315 |
#define _mm_rsqrt28_round_sd(A, B, R) \ |
| 316 |
__builtin_ia32_rsqrt28sd_round(A, B, R) |
| 317 |
|
| 318 |
#define _mm_rsqrt28_round_ss(A, B, R) \ |
| 319 |
__builtin_ia32_rsqrt28ss_round(A, B, R) |
| 320 |
|
| 321 |
#endif |
| 322 |
|
| 323 |
#define _mm512_exp2a23_pd(A) \ |
| 324 |
_mm512_exp2a23_round_pd(A, _MM_FROUND_CUR_DIRECTION) |
| 325 |
|
| 326 |
#define _mm512_mask_exp2a23_pd(W, U, A) \ |
| 327 |
_mm512_mask_exp2a23_round_pd(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 328 |
|
| 329 |
#define _mm512_maskz_exp2a23_pd(U, A) \ |
| 330 |
_mm512_maskz_exp2a23_round_pd(U, A, _MM_FROUND_CUR_DIRECTION) |
| 331 |
|
| 332 |
#define _mm512_exp2a23_ps(A) \ |
| 333 |
_mm512_exp2a23_round_ps(A, _MM_FROUND_CUR_DIRECTION) |
| 334 |
|
| 335 |
#define _mm512_mask_exp2a23_ps(W, U, A) \ |
| 336 |
_mm512_mask_exp2a23_round_ps(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 337 |
|
| 338 |
#define _mm512_maskz_exp2a23_ps(U, A) \ |
| 339 |
_mm512_maskz_exp2a23_round_ps(U, A, _MM_FROUND_CUR_DIRECTION) |
| 340 |
|
| 341 |
#define _mm512_rcp28_pd(A) \ |
| 342 |
_mm512_rcp28_round_pd(A, _MM_FROUND_CUR_DIRECTION) |
| 343 |
|
| 344 |
#define _mm512_mask_rcp28_pd(W, U, A) \ |
| 345 |
_mm512_mask_rcp28_round_pd(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 346 |
|
| 347 |
#define _mm512_maskz_rcp28_pd(U, A) \ |
| 348 |
_mm512_maskz_rcp28_round_pd(U, A, _MM_FROUND_CUR_DIRECTION) |
| 349 |
|
| 350 |
#define _mm512_rcp28_ps(A) \ |
| 351 |
_mm512_rcp28_round_ps(A, _MM_FROUND_CUR_DIRECTION) |
| 352 |
|
| 353 |
#define _mm512_mask_rcp28_ps(W, U, A) \ |
| 354 |
_mm512_mask_rcp28_round_ps(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 355 |
|
| 356 |
#define _mm512_maskz_rcp28_ps(U, A) \ |
| 357 |
_mm512_maskz_rcp28_round_ps(U, A, _MM_FROUND_CUR_DIRECTION) |
| 358 |
|
| 359 |
#define _mm512_rsqrt28_pd(A) \ |
| 360 |
_mm512_rsqrt28_round_pd(A, _MM_FROUND_CUR_DIRECTION) |
| 361 |
|
| 362 |
#define _mm512_mask_rsqrt28_pd(W, U, A) \ |
| 363 |
_mm512_mask_rsqrt28_round_pd(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 364 |
|
| 365 |
#define _mm512_maskz_rsqrt28_pd(U, A) \ |
| 366 |
_mm512_maskz_rsqrt28_round_pd(U, A, _MM_FROUND_CUR_DIRECTION) |
| 367 |
|
| 368 |
#define _mm512_rsqrt28_ps(A) \ |
| 369 |
_mm512_rsqrt28_round_ps(A, _MM_FROUND_CUR_DIRECTION) |
| 370 |
|
| 371 |
#define _mm512_mask_rsqrt28_ps(W, U, A) \ |
| 372 |
_mm512_mask_rsqrt28_round_ps(W, U, A, _MM_FROUND_CUR_DIRECTION) |
| 373 |
|
| 374 |
#define _mm512_maskz_rsqrt28_ps(U, A) \ |
| 375 |
_mm512_maskz_rsqrt28_round_ps(U, A, _MM_FROUND_CUR_DIRECTION) |
| 376 |
|
| 377 |
#define _mm_rcp28_sd(A, B) \ |
| 378 |
__builtin_ia32_rcp28sd_round(B, A, _MM_FROUND_CUR_DIRECTION) |
| 379 |
|
| 380 |
#define _mm_rcp28_ss(A, B) \ |
| 381 |
__builtin_ia32_rcp28ss_round(B, A, _MM_FROUND_CUR_DIRECTION) |
| 382 |
|
| 383 |
#define _mm_rsqrt28_sd(A, B) \ |
| 384 |
__builtin_ia32_rsqrt28sd_round(B, A, _MM_FROUND_CUR_DIRECTION) |
| 385 |
|
| 386 |
#define _mm_rsqrt28_ss(A, B) \ |
| 387 |
__builtin_ia32_rsqrt28ss_round(B, A, _MM_FROUND_CUR_DIRECTION) |
| 388 |
|
| 389 |
#ifdef __DISABLE_AVX512ER__ |
| 390 |
#undef __DISABLE_AVX512ER__ |
| 391 |
#pragma GCC pop_options |
| 392 |
#endif /* __DISABLE_AVX512ER__ */ |
| 393 |
|
| 394 |
#endif /* _AVX512ERINTRIN_H_INCLUDED */ |
